Research map / 3D vision / Human pose estimation
Generalized Feedback Loop for Joint Hand-Object Pose Estimation
Markus Oberweger, Paul Wohlhart, Vincent Lepetit
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add feedback loop (synthesizer + updater) to DeepPrior++ predictor | DeepPrior++ · NYU hand pose | average 3D error (mm): 12.2 mm -> 10.8 mm | Improved |
| use learned updater instead of direct L-BFGS-B image-difference optimization (ablation) | feedback loop · NYU hand pose | average 3D error (mm): 32.3 mm vs 12.2 mm initial pose | Improved |
| add feedback loop to hand-object predictors (ablation) | hand-object pose estimator · DexterHO | combined error (mm): 19.3mm -> 17.6mm | Improved |
| use separate CNNs for hand and object instead of a single network (ablation) | hand-object pose estimator · DexterHO | average combined error (mm): 36.6 mm -> 17.6 mm | Improved |
| use more than two updater iterations (ablation) | feedback loop · DexterHO | accuracy | No clear effect |
| train updater with Particle Swarm Optimization instead of L-BFGS-B for image-based fitting (ablation) | image-based optimization · NYU hand pose | accuracy | No clear effect |
| train predictor, synthesizer and updater jointly end-to-end from pretrained init (ablation) | feedback loop · NYU hand pose | accuracy | No clear effect |
| train updater with object synthesizer CNN instead of OpenGL renderer (ablation) | feedback loop · DexterHO | accuracy | No clear effect |
| use stride instead of max-pooling in updater CNN (ablation) | updater CNN · NYU hand pose | accuracy | Improved |
| train object localizer on spherical objects, test on cuboids (ablation) | object localizer CNN · DexterHO | localization error (mm): 40 mm worse | Hurt |
| train hand localizer on spherical objects, test on cuboids (ablation) | hand localizer CNN · DexterHO | localization error (mm): 1 mm worse | Hurt |
| layer-wise training of synthesizer instead of end-to-end (ablation) | synthesizer CNN · NYU hand pose | reconstruction error | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Crossing Nets — Crossing Nets: Combining GANs and VAEs With a Shared Latent Space for Hand Pose Estimation (CVPR 2017) · better on 1 of 1 shared result
- Lie-X — Lie-X: Depth Image Based Articulated Object Pose Estimation, Tracking, and Action Recognition on Lie Groups (IJCV 2017) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Real-Time 3D Hand Pose Estimation with 3D Convolutional Neural Networks (TPAMI 2018 · similarity 0.92)
- 3D Convolutional Neural Networks for Efficient and Robust Hand Pose Estimation From Single Depth Images (CVPR 2017 · similarity 0.92)
- DeepHand: Robust Hand Pose Estimation by Completing a Matrix Imputed With Deep Features (CVPR 2016 · similarity 0.92)
- Robust Hand Pose Estimation During the Interaction With an Unknown Object (ICCV 2017 · similarity 0.92)
- Learning to Estimate 3D Hand Pose From Single RGB Images (ICCV 2017 · similarity 0.92)
- Hand PointNet: 3D Hand Pose Estimation Using Point Sets (CVPR 2018 · similarity 0.92)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
